Do Scientists Pay to Be Scientists?

*
By

Scott Stern, Northwestern University and NBER

November 2003

I would like to thank those biologists and postdoctoral administrators who participated in the survey for their time and
thoughtfulness. By prior consent, names of specific institutions are withheld from the final analysis. John Abowd, Bob
Gibbons, Joshua Gans, Rebecca Henderson, Richard Nelson, Eric von Hippel, Lynne Zucker, seminar participants at
numerous universities, an associate editor and three extremely helpful anonymous referees provided useful suggestions.
Kwanghui Lim and Eugene Orlov, in conjunction with undergraduate research assistants Amalia Miller, Vanessa Bhark,
Allison Neizmik, and Sarah Huber, provided extraordinary research assistance. This research was completed while a
Health and Aging Fellow at the National Bureau of Economic Research, whose hospitality is gratefully acknowledged.
All errors and omissions are my own.

Do Scientists Pay to Be Scientists?

ABSTRACT

This paper explores the relationship between wages and the scientific orientation of R&D
organizations. Firms who adopt a science-oriented research approach (i.e., Science) allow
their researchers to pursue and publish an individual research agenda.. The adoption of Science
may be associated with a taste for Science on the part of researchers (a Preference effect)
and/or as a ticket of admission to gain earlier access to scientific discoveries with commercial
application (a Productivity effect). These two effects differ in their impact on wages. Whereas
the Preference effect contributes to a negative compensating differential, the Productivity effect
may result in rent-sharing. However, since Science may be adopted by firms employing higherquality researchers, cross-sectional evaluations of wages and Science may be biased by
unobserved heterogeneity. To overcome this bias, this paper introduces a novel empirical
approach. Specifically, prior to accepting a given job, many scientists receive multiple job
offers, allowing for the calculation of the wage-Science relationship, controlling for differences
in salary levels offered to individual researchers. Using a dataset composed of multiple job
offers to postdoctoral biologists, the results suggest a negative relationship between wages and
Science. These findings are robust to restricting the sample to non-academic job offers, but the
findings depend critically on the inclusion of researcher fixed effects. Conditional on perceived
ability, scientists do indeed pay to be scientists.
Journal of Economic Literature Classification: Nonwage Labor Costs and Benefits (J32);
Professional Labor Markets and Occupations (J44); Industry Studies, Drugs (L65); Innovation
and Invention (O31); Management of R&D (O32). Keywords: Economics of science, R&D
employment, R&D management, multiple job offers, pharmaceuticals, drug discovery, biology,
hedonic wage equation, compensating differentials, rent-sharing.

I.

Introduction
Since the seminal work of Nelson (1959) and Arrow (1962), economists and management

scholars have attempted to understand the incentives for abstract knowledge production. To the
extent that abstract knowledge serves as a non-rivalrous input into technological innovation, its
production plays a critical role in the process of economic growth (Romer, 1990). Because
knowledge production is costly to monitor and subject to expropriation, the level of production
may be inefficiently low in the absence of alternative institutions.
On the one hand, several institutions supporting the production of knowledge (from trade
secrecy to prizes to patents) have been identified and investigated, both theoretically and
empirically (Wright, 1983; Levin et al, 1987). However, the economic consequences of the
institution most closely associated with the production of abstract knowledge Science have
only recently begun to be explored (David and Dasgupta, 1994). Science differs from alternative
knowledge production systems insofar as researchers are offered substantial discretion in
choosing research projects and the reward structure is premised on establishing intellectual
priority (i.e., being first to make a discovery) through journal publication.
This paper considers two explanations for the use of a Science-oriented approach. First,
researchers may have preferences for interacting with discipline-specific scientific communities
and for receiving recognition for their discoveries. Simply put, scientists may have a taste for
Science (Merton, 1973; David and Dasgupta, 1994). Whether such preferences are intrinsic or
reflect long-term career motivations, a scientific ethic is critical for the effectiveness of
Science as an economic institution. A second motivation for participation in Science,
particularly for private firms, may be the productivity benefits in terms of technological
innovation. Firms who adopt Science may gain earlier and more detailed access to new
discoveries and so may be purchasing a ticket of admission which pays itself off in terms of
higher R&D productivity and a higher rate of technological innovation (Cohen and Levinthal,
1990; Rosenberg, 1990).
At one level, there is no inherent conflict between these two perspectives on the drivers
of the adoption of a scientific orientation. Scientists may value participation in Science (referred
to as the Preference effect), and, at the same time, firms may participate to capture spillovers
(referred to as the Productivity effect). However, these perspectives do offer competing
1

economic implications, most notably for the employment relationship: whereas the Preference
effect contributes to a negative compensating differential between Science and wages, the
Productivity effect raises the possibility of rent-sharing between firms and researchers (and so a
positive association between wages and Science). As well, the relationship between wages and
Science might reflect a skill bias: under the winner-take-all nature of the scientific reward
system, the expected benefits to Science will be higher for higher-ability researchers. As a
result, firms who employ scientists of higher ability will tend to adopt Science as well.
After reviewing the new economics of science in Section II, Section III develops this
argument by presenting a simple economic model incorporating the Preference effect, the
Productivity effect, and a distribution of talent (or perceived ability) among scientists. The
model incorporates the potential for rent-sharing among firms and scientists. A key result stands
out: the relationship between wages and Science depends on the relative salience of the
Preference and Productivity effects, and prior empirical investigations may have confounded the
Preference effect, the Productivity effect, and the impact of unobserved heterogeneity among
scientists.
This theoretical framework motivates the development and implementation of a novel
empirical methodology (Section IV) that controls for the presence of unobserved heterogeneity.
Specifically, prior to accepting a specific offer, many professionals receive multiple job offers.
Each offer is composed of a wage offer and job characteristics (observed to the employee).
Since a formal job offers confer a legal responsibility on the firm, offers will not be made unless
the firm is willing to employ the worker, and firms will not make offers with a zero probability
of acceptance. Using a sample of individuals seeking their first full-time employment, we are
able to calculate wage-Science combinations for randomly selected workers at a point in time.
By surveying candidates about detailed characteristics of their multiple job offers (e.g., Does
this job give you permission to publish in the scientific literature?), it is possible to estimate the
relationship between wages and Science, after taking individual heterogeneity into account
through the inclusion of person-specific fixed effects. As such, the results reported here likely
overcome the bias associated with earlier work in both labor economics and management which

examines cross-sections of employees or focuses attention on job-switching behavior.1

This survey-based methodology is applied to a sample of biologists who are (just)
completing a job search. PhD biologists (e.g., those completing their first postdoctoral
fellowship at a top-tier medical center or university) participate in a (moderately formal) job
market in which they attempt to garner long-term employment. While we can control for
differences among jobs which do not relate to their scientific orientation (such as the impact on
their career, or whether the job is in a start-up firm), the analysis is conditioned on a set of
research-oriented job offers, all of which use the candidates specialized biology background.
However, the jobs differ in terms of the degree to which the candidate can participate in Science,
either through the publication process or through the ability to choose (or continue to work on)
an individual research agenda. Critically, the job market for biologists is synchronized enough
so that, prior to accepting a job offer, some candidates do indeed receive multiple offers.
While the sample size is small, the data point to several important findings. Most
importantly, there seems to be a tradeoff between offered wages and the scientific orientation of
firms. Offers which contain science-oriented provisions, ranging from permission (or incentives)
to publish in the scientific literature, to the flexibility to choose or continue research projects, are
associated with lower monetary compensation and starting wages. These results are robust to
several different types of controls (including job type) as well as restricting the sample to those
job offers from the non-academic sector. In addition, the data suggest that accepted and rejected
job offers are comparable to one another (consistent with an equilibrium theory of hedonic wage
determination), and job candidates receiving multiple offers are not appreciably different from
candidates receiviong only one offer. However, the findings depend critically on controlling for
individual heterogeneity; simple cross-sectional comparisons result in a significant upward bias
and an underestimate of the size of the compensating differential for Science.
Overall, the contribution of this paper arises from (a) developing a simple model of the
relationship between wages, productivity, and participation in Science based on insights from the
new economics of Science, (b) estimating the relationship between wages and Science using a
novel empirical methodology, and therefore (c) providing an assessment of the relative salience

Rather than reviewing the labor economic literature here, I defer discussion of this literature until Section IV.

of researcher preferences and firm incentives, controlling for differences among individual
researchers. As an economic institution, Science helps overcome the inability to provide
incentives for the production and diffusion of abstract knowledge: simply put, we rely on
scientists to behave in a way which values knowledge production and diffusion for its own sake.

II.

The New Economics of Science

Science is a distinctive institution in several ways: as a knowledge production system, as

an input into technological innovation, and as a reward system. The first distinguishing aspect of
Science is that it produces two extremely specialized types of knowledge: potentially testable
theories and empirical tests of these theories (Kuhn, 1963). While the contribution to the theorytesting dynamic is the primary criterion by which scientists evaluate the value of new
knowledge, Science serves a second function as a knowledge stock upon which firms draw for
technological innovation. The knowledge and techniques produced by Science often have
commercial applications which are unrelated to the initial theory-testing motivation which
spurred their development (Rosenberg, 1974). In other words, Science is an important source of
knowledge spillovers. This uneasy relationship between Science as a knowledge production
process and as an input into technological innovation is mediated by the third characteristic
feature of Science: the priority-based reward system. To receive credit for the intellectual
priority of their scientific discoveries, scientists publicize their findings as quickly as possible
and retain no formal intellectual property over their ideas (Merton, 1957; Dasgupta and David,
1994).
Each of these features helps to identify Science as an economic institution that is distinct
from a commercially-motivated knowledge production system. For example, while most
economic analysis of knowledge production emphasizes the importance of institutions such as
intellectual property, trade secrecy or entry barriers in ensuring the incentives for innovation
(Nelson, 1959; Arrow, 1962; Levin et al, 1987; Kremer, 1998), the scientific incentive system
specifically rejects non-disclosure, tight control over intellectual property, or monopolization in
the use of novel scientific knowledge (Dasgupta and David, 1994).2 I therefore define

One might also distinguish Science by its locus in universities, its association with general human capital

participation in Science to be distinguishable from commercially motivated knowledge

production, involve the formulation and testing of theories (which may but need not result in
commercial spillovers), and result in public disclosure in academic journals.3
Perhaps surprisingly, formal analysis of the priority-based reward system in economics
and management is recent.4 Two recent research streams have begun to address the distinct
economic and strategic issues associated with Science, particularly the adoption of Science by
profit-oriented firms. In the first stream, David and Dasgupta (1987; 1994) and David (1998)
evaluate how the assumption that scientists (as a group) are attempting to maximize the flow of
new knowledge influences the economic efficiency of the priority-based reward system.
Dasgupta and David argue that, relative to a system in which knowledge production is rewarded
purely by monetary rents, the priority-based reward system is relatively efficient. On the one
hand, the priority-based system discourages shirking, since lower effort is associated with lower
probability of reward. On the other hand, the system encourages knowledge diffusion, since
scientists expend effort publicizing their results in order to achieve as much credit as possible.
Given that scientists are attempting to maximize the rate of production and diffusion of scientific
knowledge (i.e., scientists have a taste for Science), this priority-based system is an effective
mechanism for achieving these goals.5 Though this argument implicitly relies on a taste for
Science by researchers, Dasgupta and David neither delineate the testable empirical implications
of this theory nor compare it to alternative theories of the economic impact of Science.
In contrast to Dasgupta and David, who evaluate the efficiency properties of Science
from the perspective of researchers having a taste for Science, Rosenberg (1990), Cohen and

investments and training, and its long-term responsiveness to overall societal needs. For present purposes, I focus on
the salient elements distinguishing Science from more traditional economic models of the research process.
3
Obviously, this is a stylized conceptualization of Science as a knowledge production system (Latour and Wollgar,
1986). While acknowledging that Science is more subtle and dynamic than proposed here, this simple
characterization allows for the development of the key hypotheses directly related to the underlying phenomena.
4
Of course, economists have studied scientists behavior to evaluate other aspects of behavior (e.g., Freemans
seminal studies of scientific labor markets (1976) or Levin and Stephans careful test of Beckers life-cycle human
capital model (1991)). See Stephan (1996) for an extremely informative synthesis of this work.
5
It is useful to note that a taste for Science is likely concentrated among individuals in research-oriented careers,
perhaps reflecting their education, career incentives or the selection of individuals in this profession. In this sense,
this study examines a more specific job attribute than traditional on-the-job amenities such as safety or pension
benefits (see Hamermesh (1984) or Viscusi (1993) for reviews). As well, David and Dasgupta (and David (1998))
also highlight important inefficiencies associated with Science, most notably the possibility of overinvestment and
overnarrowness in research due to racing effects.

Levinthal (1989, 1990), and Arora and Gambardella (1994) examine how profit-maximizing
firms might exploit Science for their own purposes. Specifically, given that Science tends to
produce knowledge spillovers, relying on Science may provide access to new discoveries,
improve the productivity of technological search, and employ codified forms of knowledge
which allow for cumulative progress (Fleming and Sorenson, 2003; Foray, 2003). Thus, private
firms who would like to exploit novel scientific knowledge must purchase a ticket of
admission which pays itself off in terms of higher R&D productivity and a higher rate of
technological innovation (Rosenberg, 1990).6
These two perspectives from the new economics of science are not mutually exclusive.
Scientists may possess a taste for participating in Science (the Preference hypothesis), and some
firms may find it worthwhile to participate in Science to capture knowledge spillovers (the
Productivity hypothesis).7 However, these effects have separate empirical implications, most
notably for the employment relationship. While the Preference effect contributes to a negative
compensating differential), the Productivity effect can contribute to a positive association
between Science and wages (as long as there is rent-sharing between firms and researchers).
Though not mutually exclusive, prior empirical work that has focused on the relationship
between Science and firm performance has confounded the Preference and Productivity effects
(e.g., in the context of the life sciences industries, see Gambardella, 1995; Powell, 1996; Zucker
and Darby, 1996; 1998, and Cockburn, Henderson, and Stern, 1999). However, similar to this
paper, recent studies highlight the tension within public/private scientific networks between
scientific advance, commercial gains, and career concerns (Owen-Smith and Powell, 2001;
Gittelman and Kogut, 2003; Murray, 2003; Lim, 2003; Murray and Stern, 2003).8
It is important to note that estimating the economic impact of Science is confounded by
the selection of more able scientists into science-oriented firms. Correlation between Science
6

More precisely, if a commercial prize is available to the first firm that translates a scientific advance into a novel
technology, then firms have incentives to gain early access to absorb novel discoveries; according to many
observers, participation in Science by at least some employees may be an efficient way to achieve this goal (Hicks,
1995).
7
A fully specified model might place limits on the share of activity attributable to the profit-maximizers, insofar as
this group may tend to exhibit free-riding in the absence of intrinsic preferences. Theoretical conditions for the
existence of a priority-based equilibrium is an interesting area for future work but beyond the scope of this paper.
8
Indeed, these studies reflect a broader body of research on the development of local technical communities, also

and researcher ability arises from the internal logic of the priority-based reward system. First,
higher-quality researchers may be willing to trade off more income to earn the higher expected
prestige rewards associated with their ability. Second, firms who employ higher-quality
scientists may find participating in Science more attractive; with their better reputations, these
researchers will gain better access to the external scientific community, (Zucker and Darby,
1996).9 Consequently, without detailed controls for scientist ability, empirical assessments of
the impact of Science on performance may be biased upwards since Science may be associated
with the higher performance associated with higher-ability researcher distributions.10

III.

A Simple Economic Model of Science and Wages

We now build on this qualitative discussion and develop a simple model incorporating

the Preference effect, the Productivity effect, a distribution of talent among scientists, and rentsharing between firms and scientists. By deriving the equilibrium relationship between Science
and wages, we identify the empirical implications of the Preference and Productivity effects.
The model is composed of two stages. In the first stage, firm j chooses whether to adopt
a scientific orientation for its R&D department (SCI = 1, else SCI = 0).11 In the second stage,
firms hire a single researcher with ability i (observable to market participants but unobserved by
the econometrician). As noted earlier, the population of researchers is subject to a skewed
distribution of talent and so this variance is directly incorporated into the model. For firm j, the
quality of worker i (the most attractive scientist who applies for a job at firm j) is drawn from a

referred to as communities of practice (Feldman, 1994; Rosenkopf and Tushman, 1998; Almeida and Kogut, 1999).
9
Zucker and Darby (1996) and Zucker, Darby and Brewer (1998) present evidence that biotechnology firms are
more successful if they employ or interact (through copublication) with star researchers. While this evidence
highlights the importance of the human capital distribution, they do not explicitly distinguish the underlying
economic forces leading to such a bias or the impact of these structural characteristics on economic observables.
10
Of course, other forces may also determine the adoption of Science. For example, firms may differ in their
idiosyncratic return to Science as a result of their therapeutic specialization (Cockburn, Henderson and Stern, 1999).
11
As discussed in Section II, participation in Science involves doing research which involves the formulation and
testing of theories and results in public disclosure of findings through the academic publication process; however,
science-oriented firms, in achieving the goals of Science, will also tend to (a) allow researchers discretion in
choosing new research projects or continuing old ones and (b) tend to base promotion decisions on a researchers
external scientific reputation. The model does not distinguish between these characteristics. As well, the model
assumes that each firm chooses a firm-wide scientific orientation (rather than tailoring the scientific orientation to
each employee), an assumption consistent with prior qualitative and quantitative evidence (Gambardella, 1994;
Zucker and Darby, 1996; Henderson, Cockburn, and Stern, 1999).

firm-specific distribution, g j ( ) , bounded below at zero and with mean j .12 Each scientists
utility depends on the offered wage and the preference for a science-oriented job environment:13

U i = 0 + S i SCI j + w j

(1)

Scientists of higher ability place higher value on a science-oriented research environment; as

discussed earlier, this interaction is strongly implied by the internal logic of the priority-based
reward system and the ever-increasing rewards from prestige.14 Firms, on the other hand,
earn profits according to the ability of hired scientists, the wages paid to these employees ( w j ),
and their scientific orientation:

i , j = i ( 0 + S SCI j ) wi , j SCI j

(2)

While firms pay a fixed fee to adopt a scientific orientation,15 the benefits that the firm
receives from adopting a scientific orientation depend on the quality of the scientist. Like our
assumption about the interaction in (1), the interaction in (2) is motivated by the nature of the
priority-based reward system: higher-quality scientists will have better access to the external
scientific community and will be able to monitor external developments more skillfully than
less-talented colleagues. In other words, the benefits from Science are skill-biased.
Firms who adopt Science earn a quasi-rent, which is increasing in i. However, if the
firm faces a search cost for new scientists and scientists can (credibly) threaten to receive
additional job offers from other science-oriented firms, then scientists may extract some of this
quasi-rent in wage bargaining.16 To account for this possibility, I follow recent work on rent-

12

Alternatively, one could assume that the firm employs N researchers of known quality with mean ability

13

j.

To highlight the relationship between the wages from research jobs and scientific orientation, I adopt Rosens
(1974, 1986) hedonic characteristics approach and further assume that (a) each scientist supplies one unit of labor
inelastically, (b) there is a competitive (no search cost) labor market for jobs where SCI=0, (c) all firms observe the
same information about each scientist, (d) firms cannot verify the characteristics of competing job offers (until they
are accepted), and (e) except for differences in talent, all scientists share the same utility function. As such, the
model abstracts away from some of the more subtle issues which arise in the context of hedonic wage determination
when workers preferences for nonpecuniary characteristics are heterogeneous (Hwang, et al, 1992), or the
nonpecuniary characteristic has continuous support, all jobs require substantial search, and firms differ in the cost of
providing the nonpecuniary benefit (Hwang, et al, 1998).
14
As well, this interaction could represent a reduced-form income effect (Weiss, 1976; Sattinger, 1977).
15
One could imagine as the per-scientist cost of sending scientists to conferences, financing a staff to approve
scientific articles, or the budget for discretionary research activities (e.g., specialized materials or equipment).
16
More precisely, each scientists bargaining position depends on her (exogenous) outside option in a non-scienceoriented jobs, the average expected cost of searching for another science-oriented offer and the expected cost to the

sharing and incorporate a rent-splitting parameter, (0,1) , which determines the allocation
of the quasi-rent between scientist and firm (Abowd and Lemieux, 1993; van Reenen, 1996).
As a result, the second stage wage equilibrium is given by:

wi*, j = i 0 + i ( S S ) SCI j

(3)

Equation (3) describes the empirical relationship between wages and scientific
orientation, conditional on the quality of scientists. As long as the compensating differential
parameter, S, is larger (smaller) than that part of the quasi-rent extracted by the scientist, then
wages will be decreasing (increasing) in the firms scientific orientation. As well, it is important
to note that, by the very logic of the Preference and Productivity effects, the labor market for
researchers will straddle both academic and non-academic employment opportunities, and so
offers from both sectors will ultimately identify the relationship between Science and offered
wages. Of course, in our empirical work, it will be important to distinguish between scienceoriented job characteristics per se and the more diffuse set of attributes associated with
university employment. As well, the adoption of Science will be concentrated in those firms
who expect to recruit high quality scientists (i.e., SCI j = 1 iff j >

= * ); both the
(1 ) S + S

Preference and Productivity theories contribute to a positive impact of Science on measures of

firm performance.

Finally, it is useful to highlight that the model has been tailored to

highlight the different economic consequences of the Preference and Productivity effects. It is of
course possible that the Preference effect does not capture an intrinsic utility parameter but
instead reflects career concerns on the part of researchers. Career concerns are particularly
salient since scientific publications are a publicly observable signal of researcher quality. It is
important to emphasize that whether the Preference effect is driven by an intrinsic concerns or
career concerns, both are alternatives to the Productivity hypothesis so far the dominant
explanation offered in the economics of technological change for the adoption of Science by
private firms.

firm of finding another job candidate of similar quality. The presence of these competing sources of bargaining
power can yield a rent-sharing equilibrium (Mortensen, 1990; Abowd and Lemieux, 1993; Hwang, et al, 1998).

IV.

An Empirical Model of Science and Wages

The wage equation is a function of the Productivity and the Preference parameters, S

and S; consistent estimation of (3) therefore provides insight into the precise motivation for and
welfare implications of the adoption of Science. Both parameters, S and S exert competing
effects on w* . As such, examining the labor market for scientists seems like a fruitful place to
begin distinguishing the relative importance of these two effects.17 However, a central issue for
empirical work is that the potential for bias from heterogeneity in worker ability is likely to be
important in the current context. Not only is the adoption of Science predicted to covary with
researcher ability according to the underlying theory, but the distribution of scientific ability is
well-known to be skewed (Stephan, 1996). This paper addresses this problem by introducing a
novel empirical methodology exploiting the fact that in job markets for novice professionals
(i.e., no prior career-oriented job experience), many candidates receive multiple job offers prior
to accepting a single job offer. By observing more than one combination of wages and job
characteristics for each of these novice job candidates, we are able to construct different points
on the wage-characteristics curve for a randomly selected worker at a point in time.18 By
exploiting the specific institutions of professional labor markets, this methodology likely leads to
a substantial reduction in the bias associated with unobserved heterogeneity.
Consider an empirical wage equation that does not account for ability heterogeneity:

wi , j = 0 + S SCI j + i , j

(6)

If ability, i, is uncorrelated with the adoption of SCI, then E (S ) = ( S S ) , or the relative

salience of the preference versus productivity effect, evaluated at the mean ability level.
However, if SCI is associated with higher-quality researchers, then there will be an upward bias
on the estimate of the marginal impact of Science on wages. This bias combines two sources of
bias identified in earlier applications. On the one hand, higher-ability individuals will tend to
consume higher levels of positively valued hedonic characteristics. While earlier work has

17

I am not aware of any earlier empirical studies in the economics of science which have attempted to disentangle
these parameters, or the relative importance of the talent effect in biasing cross-sectional results.
18
For this condition to hold, it is of course important that individuals who receive multiple offers are not
systematically different than those who receive only one offer. This is ultimately an empirical question; in Section
VI, we provide evidence suggesting that the offers received by these two groups are in fact comparable.

10

attributed such covariation to wealth effects (Weiss, 1976; Sattinger, 1977; Rosen, 1986) or the
impact of heterogeneous preferences (Hwang et al, 1992), the current application motivates this
bias from the nature of a prestige-based reward system. On the other hand, the potential for
skill bias from new technologies and the associated bias for productivity studies is wellestablished (Berman, Bound, and Griliches, 1994; Autor, Katz, and Krueger, 1998; Entorf and
Kramarz, 1998). Though Science is not a technology, it is an organizational practice more
likely to be adopted when higher-quality workers are employed.
This paper proposes a novel field-based methodology based on the concept of multiple
job offers -- to overcome this bias.19 This methodology is based on the process by which
professional labor markets operate, which likely reduces the above-mentioned biases. For most
graduate or postdoctoral scientific researchers, a job search involves sending out resumes to a
large number of universities and firms, receiving a smaller number of interviews with those
firms, and then receiving one or more job offers before accepting a final offer of employment.20
I exploit two facts about this process. First, since all graduates must engage in some job search,
graduating professionals are a particularly attractive sample from the perspective of selectivity.
Second, and more importantly, prior to accepting a specific offer, many professionals receive
multiple job offers. Each offer is composed of a wage offer and job characteristics (including
scientific orientation) which are observed to the employee.21 By surveying candidates about
their multiple job offers, it is therefore possible to estimate the relationship between wages and
scientific orientation, controlling for a fixed effect for each individual in the sample. In other

19

Prior research has addressed this bias through the use of ability-associated controls and the exploitation of panel
data. While the use of control variables is useful when workers are easily distinguished by observables such as
educational attainment (see, e.g., Hamermesh, 1984; Rosen, 1986; Kostiuk, 1990; Viscusi, 1993), this approach will
be of limited value if most demographics are unlikely to distinguish individuals (e.g., when there is no variation in
educational attainment among PhD recipients). As well, it is possible to exploit panel data by including a fixed
effect for each individual and examining the impact of job switches (Brown, 1980; Duncan and Holmlund, 1983;
Murphy and Topel, 1987; Abowd, Kramarz, and Margolis, 1999; Entorf and Kramarz, 1998). However, as
discussed by Gibbons and Katz (1992), job switching is (generally) endogenous and reflects a population of
individuals who are badly matched for their current positions, leading to a separate bias.
20
As mentioned earlier, the very logic of the argument suggests that the labor market will straddle both academic
and non-academic employment opportunities. However, we are careful to distinguish between the impact of Science
and a more general impact of university employment.
21
Since a formal job offer confers a legal responsibility on the firm, such offers will not be made unless the firm is
willing to employ the worker under the proposed package. As well, in equilibrium, firms will not make offers which
have a zero probability of acceptance.

11

words, the multiple job offers methodology allows us to observe different points on the wageScience curve conditional on the quality of the researcher ( wi , j = i + S SCI j + i , j ). The use of
an individual fixed effect allows us to fully take into account heterogeneity among individuals in
terms of their overall ability or their attractiveness to employers. Without such a precise control,
it would be impossible to distinguish whether science-oriented jobs are associated with a
discount (or premium) from an unobserved correlation between researcher quality and the firms
scientific orientation.
The multiple job offers methodology rests on several important assumptions. First, the
method assumes that observed job offers are comparable in terms of the seriousness of the
offer. As much as possible, the data were gathered just prior to the time the candidate accepted
an offer; according to survey respondents, these final-round offers reflected their beliefs about
the job characteristics they would be accepting if they chose an offer. As well, we check for this
effect directly by including an accepted job dummy variable into the hedonic wage regression,
providing information about whether accepted jobs are associated with significantly higher
wages as well as providing a control for the potential relationship between the Science variables
and the seriousness of the offer. Second, candidates who receive multiple job offers must be
drawn independently from the distribution of candidates. In particular, individuals with more
offers may be drawn from a more attractive portion of the distribution (Gibbons and Katz, 1992).
At one level, this is an empirical question that can be checked directly by comparing the wage
offers and characteristics of multiple offer candidates and single offer candidates. In addition,
discussions with survey respondents suggests that the receipt of multiple job offers is more likely
associated with the degree to which a candidate synchronizes their job search over all potential
employers rather than necessarily be associated with the level of ability. Third, the method
assumes that scientific orientation is uncorrelated with alternative unobserved sources of
variation in R&D productivity. If firms scientific orientation is correlated with unobserved
practices which are the structural sources of productivity gains, then the hedonic equation will
obviously be subject to omitted variable bias. However, since the survey used to implement the
procedure can be tailored to the specific labor market under study (and after detailed informal
information-gathering and piloting), it is possible to specifically ask each respondent about the
12

most visible candidates for such bias. For example, it is possible to ask detailed questions
about the possibility of promotion or career advancement at each job (another potential reason
why researchers might trade off initial wages), and so at least provide a first-order control for
this competing effect. Finally, this procedure assumes that differences in information among
offering firms about the quality and preference of candidates is uncorrelated with scientific
orientation. If science-oriented firms can judge researchers abilities more accurately than others,
then science-oriented firms may tend to make fewer but more attractive job offers, introducing a
positive bias into the hedonic estimate.

V.

Data
The data used to evaluate the hedonic wage equation for biologists was collected using an

author-developed survey administered to life science researchers.22 The survey records

information about the experience, preferences, and decisions of individual candidates on the job
market. This section describes the sample selection procedure, and then reviews the survey and
the summary statistics for the dataset.
Sample Selection
The sample population was drawn from individuals who, with some probability, were in a
position to receive multiple private sector life sciences research job offers. To ensure
compatibility across responses, the sample is restricted to PhD biologists (though several
candidates also hold an MD). The target population was therefore current researchers who held
a PhD in biology, were currently searching for a permanent research position, and expressed
interest in receiving offers from non-academic employers. To access this target population,
surveys were distributed to the following populations:
*

current postdoctoral researchers whose funding was expiring at four American

research institutions,

participants in two AAAS-sponsored Biology Job Fairs (held in Cambridge, MA

and Palo Alto, CA), and

22

Data was gathered between June, 1998, and June, 1999 by mail and telephone interviews (and a small number of
e-mail responses). Except for a small number of individual cases, the surveys were administered prior to the time the
individual commenced working for the accepted employment offer. The survey appears in Stern (1999).

13

post-PhD biologists with resumes posted to www.biomednet.com

For each data source, several candidates were interviewed more extensively in order to
ensure that each group was comparable with the others (conditional on researcher quality).
Overall, data was collected from 107 biologists who received a total of 223 job offers. Nonresearch-oriented job offers (e.g., management consulting, software start-up management, lab
management) were excluded as were observations composed of candidates who received only
one job offer. Once these jobs are excluded, the sample includes 63 job candidates who received
a total of 164 offers. It is important to emphasize that this final dataset is quite special, in that it
conditions on a set of postdoctoral biologists, most from top-tier institutions, who received
multiple job offers. However, by their demographics and average salary offers, this sample is
likely comparable to the researchers considered in other studies focused on life science
researchers who are at the margin between academic and non-academic employment (Cockburn
and Henderson, 1998; Zucker and Darby, 1996; Owen-Smith and Powell, 2001).
The Survey Instrument
The data are drawn from the information collected on the surveys administered to the
populations described above. The goal in developing the survey was to collect information
about the relationship between the job market environment and different elements of the job
offers received by life science researchers.23 The survey is composed of five sections (see Stern,
1999). The first elicits resume information about the respondents background and
demographics. Part II gathers data about the length and outcome of the job search. In Part III,
respondents compare job offers according a number of distinct dimensions and provide an
ordinal ranking of their offers. In other words, if an individual received three offers, she would
rank these three offers in terms of an individual dimension, such as Quality of Internal Research
Environment, with the highest quality offer receiving a 1, the next preferred offer a 2 and
the lowest-quality research environment rated a 3. Part IV asks more concrete data about

23

The final form of the survey resulted from an iterative process. In response to a pre-test of the survey, the survey
was amended using feedback from both researchers with prior experience in field-based survey work and focus
group discussions with biologists who served as pre-test subjects. My special thanks to Eric von Hippel for his
detailed comments on an early draft of the survey. From the perspective of this paper, the main changes in the
survey over time were to refine the descriptive wording of the scientific environment and to divide out individual
elements of the salary package (SALARY vs. BONUS, etc).

14

each individual offer, denoted hereafter as cardinal data (i.e., generally comparable in magnitude
and intensity). In addition to questions about measurable characteristics (such as salary or
permission to publish in the public scientific literature), a number of questions are asked
according to five-point Likert scales (with higher levels on the scale corresponding to higher
objective rankings). The final survey section asks for a ranking of the importance of different
job characteristics.24
It is useful to distinguish between the use of the ordinal and cardinal data. Whereas
ordinal information provides comparative information about monetary and non-monetary
qualities of each job, the cardinal data is composed of data from each candidate where the
information about one offer is independent of the information from other offers. Both types of
information are subject to limitations. Ordinal data, by its very nature, cannot be used to provide
a measure of the intensity of different rankings either within a given job characteristic or across
characteristics. On the other hand, cardinal data is subject to various types of measurement
error; for example, the use of Likert scales in regression analysis imposes strong assumptions
about how different responses differ from each other. However, the empirical approach
overcomes these issues in two distinct ways. First, throughout the analysis, fixed individual
effects are employed to derive the principal results; by construction, this limits the measurement
bias to differences across responses within individuals. Second, the analysis evaluates both the
ordinal and cardinal data, identifying those conclusions that are robust across data types.
Variables and Summary Statistics
The dataset is composed of 164 job offers from 66 individuals who received multiple job
offers; the average individual received 2.88 offers and 30% of the observed offers were the
offers actually accepted by the candidate (see Table 1 for variable definitions and descriptive
statistics).25 The first set of variables is based on the cardinal data drawn from the Job Offer
Record. The dependent variable is the baseline salary associated with each offer, SALARY.
24

This paper does not fully exploit the richness of the survey. We focus only on those aspects of the survey which
help to identify the relationship between Science and wages.
25
As mentioned earlier, we gathered demographic information but do not use it in the current analysis as it is
subsumed by the use of fixed effects. Briefly, most of the sample is married (71%) male (63%), and the average age
is just above 34. The average respondent applied for 15.7 jobs and received their first job offer after 4.3 months.
The maximal number of job offers is equal to 6, though more than 90% of the sample is composed of individuals
with between 2 and 4 offers. The key results are robust to the inclusion or exclusion of any specific individual.

15

The average SALARY offer is a little higher than $60,000 though the standard deviation is quite
large. Of course, SALARY only captures one dimension of the monetary compensation
package, and we check the robustness of our findings to the inclusion of controls associated with
other dimensions of earnings in the firm (particularly expectations of long-term compensation
growth), including a dummy for the presence of stock options (STOCK_DUMMY, mean = 0.36)
and long-term career opportunities (PROMOTION, mean = 3.49 rated on a 1-5 Likert scale).26
Several different cardinal variables are used to evaluate the scientific orientation of each
organization: whether researchers are allowed to continue to publish discoveries
(PERMIT_PUB), how strong the incentives for publication are (INCENT_PUB, rated on a 1-5
Likert scale), and whether researchers are allowed to continue postdoctoral research projects
(CONTINUE RESEARCH). Interestingly, while most research positions do permit researchers
to publish (the mean is over .9), there is a substantial variance among offers in terms of the
incentives to publish and just under 50% of positions allow researchers to continue their prior
research agenda. Finally, the analysis includes a Likert scale measure of access to cutting-edge
equipment (EQUIPMENT). Consistent with Hamermesh and Oster (2002), the adoption and use
of research tools may be driven either by productivity concerns (as tools) or by their
consumption value (i.e., as toys).
Each of our measures of the scientific orientation of the firm is imperfect, and each is
also correlated with the others (see Table A1). We therefore construct a composite index equal
to the principal factor of PERMIT_PUB, CONTINUE RESEARCH, and INCENT_PUB. This
variable, SCIENCE INDEX, is just a linear combination of the three raw measures, where the
weights are determined according to a standard principal factor analysis (see Hamilton (1992)
for a more complete description). In contrast to the raw measures, SCIENCE INDEX takes on
15 distinct values across the sample, and provides more refined variation across offers within
individuals.
Each job offer is also coded with one of six different JOBTYPE categories: Established
Firm, Start-up Firm, Government Lab, Medical Center/Hospital, University Faculty, and

26

Stern (1999) also experiments with the use of BONUS, which is equal to the signing bonus for the job offer.
BONUS is highly correlated with STOCK_DUMMY, and so we exclude BONUS from the current analysis. The
qualitative findings do not change when BONUS is included instead of STOCK_DUMMY.

16

University Postdoc (see Table A2). Though the plurality of offers are received from established
firms and start-up firms, our main analysis includes jobs from the public and university sectors.
According to our focus groups and interviews, public sector and medical center employment are
perceived to be similar to employment by private firms (the only difference being the formal
status of the organization), while academic jobs are imperfect substitutes offering job
characteristics not captured in our empirical measures (e.g., collegiality, the option value of
moving to the private sector, etc). We therefore define academic offers to be those
associated with a University Faculty or University Postdoc position, and the analysis is careful to
distinguish how the results depend on the inclusion or exclusion of offers from these categories.
The second set of variables is ordinal in nature each individual ranked their jobs in
order of their preference according to each dimension. Not surprisingly, the mean of each of
these variables is slightly less than 2 (recall that while the average number of job offers per
respondent is 2.88, the data count each job offer as a separate observation and so individuals
with a larger number of offers appear more frequently than those with only two or three offers).
The principal dependent regressor that we will use in this context is MONETARY, or the
ranking of offers in terms of overall monetary compensation. The analysis focuses on
evaluating the relationship between the MONETARY rank and two elements of the scientific
orientation of each organization: first, the overall research quality (RESEARCH QUALITY)
and whether researchers have discretion in choosing their own projects (FLEXIBILITY).27
Based on the focus group discussions and interviews, the survey also asked respondents to rank
jobs according to several control factors, such as the availability of research funding
(FUNDING), the impact of the job on career (CAREER), and the degree of fit between specific
training (JOBFIT). While this study takes each of these latter factors as a control variable, one
could imagine separate investigations of each. For example, the JOBFIT variable might provide
information about the degree of idiosyncratic bargaining power that a worker possesses with
each employee, allowing more nuanced evaluation of the degree of bargaining power held by the
employee. As well, by including CAREER in all regressions, the analysis explicitly
distinguishes between participation in Science as a career-advancement strategy and intrinsic
27

The survey includes another element of the scientific orientation (Publication Incentives) but the wording of this

17

preferences for a Science-oriented research environment.

Before turning to the empirical results, two final sampling issues arise. While there are
164 job offers for which an alternative job offer for that candidate is observed (i.e., a multiple),
many surveys were only completed with either the ordinal or cardinal data but not both. As a
result, the final sample in the cardinal analysis includes 121 job offers across 52 individuals;
when we examine non-academic offers only, this reduces to 71 offers across 30 individuals. In
the ordinal analysis, 134 job offers across 51 individuals are included, and this sample is reduced
to 74 offers over 28 individuals when examining non-academic offers only. For some offers, a
subset of the control variables are missing (e.g., the respondent did not report whether stock
options were received); in these cases, a dummy variable for a missing value has been included.

VI.

Empirical Results
This section presents the empirical evidence about the relationship between measures of

monetary compensation and the scientific orientation of organizations. The analysis proceeds in
three stages, beginning with the cardinal data, followed by the ordinal (ranked) data, and then
addressing some potential limitations and concerns with the multiple job offers methodology.
Overall, the evidence is consistent with the predictions of the Preference hypothesis.
Cardinal Data
We begin in Table 2 with a simple non-parametric comparison of the variation in salary
offers by the scientific orientation of different jobs (in the spirit of Card and Sullivan 1988). For
each of our three principal measures of scientific orientation (PERMIT_PUB, CONTINUE
RESEARCH, and INCENT_PUB), we perform a t-test of differences in the means focusing
exclusively on individuals who experienced variation in that characteristic across their offers.
We first compute the average salary offer for each individual for each of two values of a given
characteristic, creating a person-specific wage-characteristic pair.28 We then compute the
average value of each pair, yielding a weighted average salary for each individual (where the
weights are implicitly equal to 0.5 for each value of the characteristic). To abstract away from

question led to a relatively low response rate of usable responses.

28
Though INCENT_PUB is a five-point Likert scale measure, we implemented the above procedure by grouping the

18

differences in the level of salary offers across individuals, we subtract each individuals
weighted average salary from each element of their wage-characteristic pair. In other words,
for each individual and characteristic value, we calculate an average deviation measuring how
salary offers associated with a given value of a characteristic differ from the average salary offer
received by that individual. The test is a simple comparison of the deviations associated with a
high value of the scientific characteristic versus the deviations associated with a low value
of the scientific characteristic.
The upper half of Table 2 reports the results of this test for the complete sample,
including academic offers. Given the limited size of the sample (and the smaller number of
instances in which offers vary in terms of a specific attribute within individuals), the degrees of
freedom for each test is modest, ranging from 8 for the PERMIT_PUB measure to 30 for the
INCENT_PUB measure. Despite this small number, equality of the deviations in salary offers
by scientific attribute is rejected at the 10% significance level for each of the three scientific
orientation measures. These deviations are quantitatively significant as well: for example, for
PERMIT_PUB, the (weighted) average for offers where publication is allowed is more than
$14,000 lower than the (weighted) average for offers where publication is restricted.
It is important to demonstrate that such differences are not simply the result of
differences between academic and non-academic job offers. As such, the lower half of Table 2
repeats the test, excluding academic offers from the sample. Equality of the means is rejected at
the 5% level for both PERMIT_PUB and INCENT_PUB, though the test for CONTINUE
RESEARCH can only reject equality at the 20% level.. As well, the quantitative size of the
average differences is still substantial, though smaller in absolute size for PERMIT_PUB and
CONTINUE RESEARCH. In other words, though there are some differences depending on the
sample, the core findings are similar. Whether academic offers are excluded or not, and
controlling for the average salary received by each individual, salary offers associated with a
low value of the scientific orientation measure are significantly higher than salary offers
associated with a high value of the same scientific orientation measure.
Tables 3 and 4 turn to regression analyses to further assess the relationship between

INCENT_PUB responses into two groups ({1, 2, 3} versus {4, 5})

19

salary offers and the scientific orientation of the firm. In particular, we are interested in (a)
contrasting results depending on whether we take account of individual differences, (b)
controlling for other factors that may impact the salary offer for a given job, (c) comparing the
salience of different measures of the scientific orientation of each job offer and (d) comparing
how the results depend on whether academic offers are included or not.
The first three columns of Table 3 examine the relationship between the (natural log of)
offered SALARY and PERMIT_PUB, using both academic and non-academic offers.29 In some
sense, this is the most direct test of the relationship between Science and wages, since the
permission to publish is the hallmark of Science as an institution. The first column of Table 3
reports the unconditional pairwise correlation between SALARY and PERMIT_PUB, which is
both quantitatively small (.03) and statistically insignificant. In sharp contrast, once we include
individual fixed effects in (3-2), the parameter estimate becomes negative and significant (-0.27),
both statistically and quantitatively. This result is robust to a variety of control structures. As
an illustration, (3-3) includes controls for the career prospects offered by each job
(PROMOTION), the value of the compensation package (STOCK DUMMY), the seriousness
of the job offer (ACCEPTED JOB) and dummies for each job category.30 Together, these
measures should capture the most obvious sources of omitted variable bias. The coefficient on
PERMIT_PUB remains negative and significant, though the estimated coefficient is
approximately 25% smaller in absolute value.
These estimates allow us to evaluate the salience of the effects identified in the economic
model. First, at the point estimate provided in (3-3), ( S S ) = .19 . In other words, at
the mean human capital level in the observed sample, the Preference effect outweighs the impact
of the Productivity effect and implies nearly a 20% discount on the wage rate for Scienceoriented firms. Second, the difference between the cross-sectional results and the fixed effects
results provides information about the size of the ability bias. Taking the results at face value,
29

While the hedonic is in the form of ln(SALARY) throughout, the results are robust to the use of SALARY level.
In most specifications, other elements of compensation and opportunities for promotion are positively associated
with offered salaries. In one sense, this contrasts with the simplest model of wage determination where one expects
compensating differential for these characteristics. However, such evidence is consistent with the presence of rentsharing: some firms are more successful and provide higher compensation to all recruits (perhaps with some
efficiency wage benefits); for present purposes, it is useful to note that the Science effects are distinguished from
30

20

the estimates rejects the pooled regression and imply that the size of the bias is more than 20%.
Figure 1 plots the distribution of the researcher fixed effects, illustrating the magnitude of
the researcher ability effects more directly. According to Figure 1, the inter-quartile distribution
ranges from more than 40% wage discount for the 25th percentile scientist to nearly a 40% wage
premium for the scientist at the 75th percentile. These results are particularly striking when one
considers that the sample conditions on a population of PhD holders. It is useful to emphasize
that these fixed effects do not capture the scientific ability of researchers per se but differences in
the attractiveness of different individuals on the labor market at the time in question.31 While the
small sample size and simple specification suggest that these results should be treated with
caution, the large magnitude suggests the existence of differences between jobs that permit
publication (or not) and between the job offers for different individuals in the sample.
The final three columns of Table 3 incorporate multiple scientific orientation measures
into the analysis. In (3-4), we include PERMIT_PUB, CONTINUE RESEARCH,
INCENT_PUB and EQUIPMENT simultaneously (along with other controls). Each of the first
three coefficients is negative (though only CONTINUE RESEARCH is significant), and
EQUIPMENT is positive and significant.32 Rather than attempt to separately disentangle the
separate impact of each of these variables,33 the final two columns of Table 3 examine their
combined effect, using SCIENCE INDEX (the first principal factor of PERMIT_PUB,
CONTINUE RESEARCH, and INCENT_PUB) as the key regressor. The coefficient on
SCIENCE INDEX is negative in both specifications, though it becomes insignificant when job
category dummies are included. As suggested by Table 2, this reduced precision is not
surprising, given the limited amount of within-researcher variation in the sample.
Relying on a sample that combines academic and non-academic offers, Table 3 provides
evidence consistent with the relative salience of the Preference effect as well as ability bias.

this pattern.
31
Unfortunately, confidentiality concerns do not allow me to track these individuals over time since the time of the
Survey to examine their realized publishing and patenting productivity. However, as emphasized by Gittelman and
Kogut (2003), applied research productivity may be uncorrelated (or even negatively correlated) with scientific
ability; if firms understand this tradeoff, the fixed effects may be capturing potential employers assessment of
candidates ability to contribute to commercially oriented research activities.
32
This finding provides some evidence in contrast to the toys evidence offered by Hamermesh and Oster (2002).
33
In most cases, two of the three measures can be separately estimated with a significant coefficient (Stern, 1999).

21

However, as we emphasized earlier, it is important to distinguish whether these results are

simply driven by differences between the academic and non-academic sectors. Table 4 therefore
employs a dataset limited to the non-academic offers of those individuals who received multiple
non-academic offers (resulting in a sample of 71 offers across 30 individuals). In both
specifications (including PERMIT_PUB and SCIENCE INDEX as the key regressors,
respectively), there is a negative and significant correlation between each scientific orientation
measure and offered salary. The magnitudes of these effects continues to be significant as well;
for example, a one-standard deviation shift in SCIENCE INDEX is associated with more than a
6% reduction in the predicted wage. These qualitative findings are robust to the inclusion or
exclusion of controls for promotion opportunities, whether stock options are offered, and the
seriousness of the job offer. As well, in contrast to the noisy results in (3-6), the coefficient on
SCIENCE INDEX remains significant (and quantitatively important) at the 10% level even after
job category dummies are included. In other words, whether we employ PERMIT_PUB or
SCIENCE INDEX as our measure of scientific orientation and focusing only on non-academic
employment offers, the labor market for life scientists reflects a wage discount for scienceoriented job offers.
Ordinal Data
We now turn to the ordinal data to highlight results from this separate part of the survey.
Of course, since one cannot concretely identify the intensity associated with an ordinal ranking,
the use of the ordinal data is not ideal. However, this type of data can be used to support the
cardinal results; in particular, we can examine whether similar patterns hold, given that nearly
half of all respondents in the ordinal dataset are not included in the cardinal dataset.
Table 5 presents a series of regressions using the MONETARY rank of each job as the
dependent variable.34 Each regression includes both a full set of individual fixed effects as well
as controls for JOBFIT, CAREER, and FUNDING (the results are robust to alternative controls

34

A more informal analysis of the ordinal data is presented in Table A3, which presents pairwise cross-tabulations
between the MONETARY rank and four other job dimensions. While the majority of the rankings for JOBFIT and
CAREER are on or just off the diagonal of the cross-tabulation with MONETARY (indicating a positive correlation
between these dimensions), rankings for RESEARCH QUALITY and FLEXIBILITY are more dispersed. In other
words, Table A3 suggests a tradeoff between the MONETARY ranking and RESEARCH QUALITY or
FLEXIBILITY. See Stern (1999) for more details.

22

structures). Beginning with the full sample of observed job offers, (5-1) and (5-3) demonstrate
that MONETARY is negatively correlated with RESEARCH QUALITY and FLEXIBILITY,
respectively.35 These results continue to hold even when we include academic job offers,
though the FLEXIBILITY coefficient is close to but not quite significant at the 10% level (see
(5-2) and (5-4)).36 Overall, these results provide additional supporting evidence for the salience
of the Preference effect in shaping the relationship between monetary compensation and Science.
Research environments that allow workers access to high-quality research colleagues and an
ability to choose their own projects tend to offer less attractive monetary compensation.
Single versus Multiple Offers
As discussed in Section IV, the multiple job offers methodology relies on several implicit
assumptions. Among the important of these is the assumption that those individuals who receive
multiple job offers are not drawn systematically from a different part of the ability distribution
than those who receive only one offer (this is required in order for us to estimate the mean value
of ( in the population). While we cannot provide dispositive evidence regarding this assumption,
we are able to gain some insight into the seriousness of the problem by comparing the offers
associated with multiple offer individuals with the characteristics of the offers of those who
accept the first job they are offered (see Table 6A). Essentially, relative to the variance
associated with each of the characteristics, the single-offer averages are not substantially
different than the multiple-offer averages (none of the means is significantly different from each
other). For example, while the mean salary offer associated with single offer candidates is 8%
lower, the likelihood of being able to publish is 6% higher. Along with the informal evidence
(drawn from detailed interviews with job market participants) that the likelihood of receiving a
multiple offer depends on whether one synchronizes the job search so that all employment offers
are made at roughly the same time, this comparison suggests that there is no obvious source of
evidence for selectivity associated with the use of the multiple job offers.
Accepted Versus Rejected Offers
35

If we take the estimates at face value, the results suggests that as the value of RESEARCH QUALITY increases
by one, the predicted value of MONETARY declines approximately .3.
36
The results continue to hold if RESEARCH QUALITY and FLEXIBILITY are included simultaneously, though
the results are noisier when we limit exclusively to non-academic offers.

23

A second implicit assumption of the methodology is that all of the observed offers are
equally serious. If, for example, firms with a science-oriented research approach pursued a
hiring strategy in which they made a larger number of less attractive offers, then the negative
correlation found in the data would be a remnant of this strategy rather than the fundamental
preference and productivity forces identified earlier. Indeed, concern over the degree of
seriousness of offers motivated the use of the ACCEPTED JOB dummy in most of the analysis;
in all of the fixed effects regressions, this variable was quantitatively small and statistically
insignificant. In other words, if one excludes the ACCEPTED JOB dummy, the residuals from
the hedonic wage equation cannot be explained with the ACCEPTED JOB dummy. In Table 6B,
we employ a more direct approach, comparing the characteristics associated with accepted and
rejected offers. While accepted offers are associated with somewhat higher salaries
(approximately 6%), they are also (weakly) associated with a scientific orientation (e.g., the
mean of PERMIT_PUB and EQUIPMENT are somewhat higher while the mean of the other
characteristics are essentially equal across the two groups). While these results do not rule out
the absence of bias, they do argue against a strong negative correlation between the seriousness
of the offer and Science. As such, it is unlikely that the negative correlation between offered
wages and scientific orientation is simply reflecting a low-ball job offer strategy by more
scientifically oriented firms.

VII. Discussion and Conclusion

Though the sample size is small, the results consistently point towards the possibility that
scientists pay a compensating differential to participate in Science. While the theoretical
relationship between Science and wages is ambiguous (depending on the Preference and
Productivity effects, and the degree of rent-sharing), the empirical evidence suggests that the
balance is tilted in favor of the Preference effect. Offers from Science-oriented firms are
associated with lower wages and a lower ranking in terms of monetary compensation. This
finding is robust to different characterizations of Science, different control structures, and
whether we limit the sample exclusively to non-academic offers. However, these results do
depend on controlling for differences among workers in terms of their ability, and cross-sectional
analysis yields an upward bias of the estimated relationship between wages and Science.
24

These findings have potential implications for the estimates of the benefits from scienceoriented practices in the context of prior R&D productivity studies. While prior research in the
economics of technical change have focused almost exclusively on the Productivity effect, such
an interpretation ignores two sources of bias: the wage-savings associated with the
compensating differential and the unmeasured correlated between the adoption of scientific
practices and average researcher ability. While prior researchers have acknowledged the
possibility of bias (Henderson and Cockburn, 1994), the results, taken at their face value, suggest
that these two effects may explain an important portion of the overall measured effect.
Indeed, there may be a conflict between the preferences of researchers to participate in
Science and the incentives for firm participation. Consistent with evidence offered by Cockburn,
Henderson and Stern (2003), Gittelman and Kogut (2003) and Murray (2003), science-oriented
firms face important challenges in providing appropriate incentives for researchers who
simultaneously engage in basic and applied research activities. Of course, these results depend
on the fact that the sample is drawn from an extreme case, and biologists with PhDs are
perhaps uniquely associated with a taste for Science. While the generality of these results
remains an open area for future research, it should be highlighted that life sciences researchers
now compose over one-third of all graduating PhDs in the natural sciences (NSF, 1999).
An additional perspective on these findings is that they suggest that a profession-specific
ethic participation in the public scientific community has real effects on economic
observables such as wages and perhaps productivity. In particular, a taste for Science is most
likely not universally shared among all workers but is concentrated among workers in researchoriented careers. Moreover, scientists are by no means the only professional community in
which profession-specific values may influence behavior. For example, if physicians claim to
value the health of their patients (as they often do), do physicians pay to cure their patients?
Specifically, do HMOs who offer higher-quality health care (in terms of fewer restrictions on
costly procedures and the like) extract a discount on the salary they offer new physicians as a
result of their pro-patient reputation? Similarly, lawyers often claim to value justice in the legal
system. Is this reflected in lower wages for those firms who participate in pro bono activities or
who are more directly involved in courtroom activity or appellate work (as opposed to corporate
law)? Finally, computer programmers have increasingly foregone current income to participate
25

in the open software movement, a choice which may reflect intrinsic preferences or career
concerns (Lerner and Tirole, 2002). Perhaps surprisingly, following the seminal study of
Friedman and Kuznets (1954), there has been almost no systematic economic analysis of the
impact of professional ethics on labor markets or economic organization (though Weisbrod
(1983) is an important exception).
This paper provides a new tool to perform these future studies.37 Similar to the current
context, accounting for human capital effects will be paramount in any professional labor
market; however, the multiple job offers methodology will also apply to these markets. The
average law student receives several job offers during their law school career, and, after
residency, many physicians receive multiple offers from different HMOs. Relative to the
number of postdoctoral biologists seeking first-time employment in the private sector, law
schools and medical residencies offer potentially much larger cohort populations from which to
gather job offer data.
Finally, the framework and empirical results may have more general implications for how
we interpret Science as an economic institution. Specifically, there is a broad (though not
unanimous) agreement that basic research of some type should be subsidized or encouraged
because of its role in spurring long-term technological change and economic growth (Romer,
1990); however, the exact form of that subsidy is much more contentious (Wright, 1983). On
the one hand, much public policy discussion argues that researchers even basic researchers
should be evaluated on their ability to long-term commercial consequence of their research
proposals (Committee for Economic Development, 1998). On the other hand, others contend
that the peer-reviewed scientific research funding and publication system operates according to
an internal logic which can be easily undermined by providing high-powered explicit incentives
for applied commercial output (David and Dasgupta, 1994; Nelson and Rosenberg, 1994). This
study points towards the potential for an empirical foundation for this latter perspective.

37

For example, Miller (1999) employs the multiple job offers methodology to evaluate the value placed on diversity
and mentoring programs by minority MBA candidates. Hsu (2003) assesses the value placed by technology
entrepreneurs on the reputation and resources of venture capitalists when choosing among multiple VC offers.
Though not profession-specific, these studies highlight how this methodology can be extended to other settings. In a
somewhat similar vein, Royalty (1999) exploits the observation of multiple health insurance options by individual
workers in a given firm to directly estimate the preference parameters.

26

Specifically, if the producers of abstract knowledge (a long-term public good) are sensitive to the
integrity and prestige associated with the production of pure knowledge, then society may be
able to produce such knowledge at lower cost than would be the case if researchers were only
sensitive to the tradeoff between effort and realized income. Because letting the scientific
community establish its own internal rules for prestige and recognition may reduce the cost of
knowledge production, total spillovers from knowledge production into technological innovation
may depend on the degree of insulation from commercial incentives. Exploring the role that the
institutions of Science play in shaping R&D productivity therefore seems like a promising area
for further study in management and economics.

27

REFERENCES
Abowd, J. and Lemieux, T. (1993). The Effects of Product Market Competition on Collective Bargaining
Agreements: The Case of Foreign Competition in Canada, Quarterly Journal of Economics, 108 pp. 9831014.
Abowd, J., Kramarz, F., and D. Margolis (1999). High Wage Workers and High Wage Firms, Econometrica 67(2)
pp. 251-333.
Abrevaya, J. and J. Hausman (1999). Semiparametric Estimation with Mismeasured Dependent Variables: An
Application to Duration Models for Unemployment Spells, mimeo, MIT Department of Economics.
Almeida, P. and B. Kogut (1999). Localization of Knowledge and the Mobility of Engineers in Regional
Networks, Management Science, 45(7), pp. 905-918.
Arora, A. and A. Gambardella (1993). Evaluating Technological Information and Using It, Journal of Economic
Behavior and Organization, 24(1) pp. 91-114.
Arrow, K. (1962). Economics of Welfare and the Allocation of Resources for Invention, The Rate and Direction
of Inventive Activity ed. R.R. Nelson Princeton: Princeton University Press.
Autor, D.H., Katz, L.F. and A.B. Krueger (1998). Computing Inequality: Have Computers Changed the Labor
Market? Quarterly Journal of Economics, 113(4), pp. 1169-1214.
Berman, E., Bound, J. and Z. Griliches (1994). Changes in the Demand for Skilled Labor in U.S. Manufacturing
Industry: Evidence from the Annual Survey of Manufacturing, Quarterly Journal of Economics, 109, pp.
367-397.
Brown, C. (1980). Equalizing Differences in the Labor Market, Quarterly Journal of Economics, 94, pp. 113-134.
Card, D. and D. Sullivan (1988). Measuring the Effect of Subsidized Training Programs on Movement in and Out
of Employment, Econometrica 56, pp. 497-530.
Cockburn, I. and R. Henderson (1998). Absorptive Capacity, Coauthoring Behavior, and the Organization of
Research in Drug Discovery. Journal of Industrial Economics, 46(2), pp. 157-182.
Cockburn, I., Henderson, R. and S. Stern (1999). The Diffusion of Science-Driven Drug Discovery: Organizational
Change in Pharmaceutical Research, NBER WP 7359.
Cohen, W. and R. Levin (1989). Empirical Studies of Innovation and Market Structure, Handbook of Industrial
Organization eds. R. Schmalensee and R. Willig. Amsterdam: Elsevier.
Cohen, W. and D. Levinthal (1989). Innovation and Learning: The Two Faces of R&D, Economic Journal,
99(397), pp. 569-596.
Cohen, W. and D. Levinthal (1990). Absorptive Capacity: A New Perspective on Learning and Innovation,"
Administrative Science Quarterly,. 35, pp. 128.152.
Cole, J. and S. Cole (1973). Social Stratification in Science. University of Chicago Press.
Committee for Economic Development (1998). Americas Basic Research: Prosperity Through Discovery.
Crane, D. (1972). Invisible Colleges. Diffusion of Knowledge in Scientific Communities, Chicago: University of
Chicago Press.
Dasgupta, P. and P. David (1987). Information Disclosure and the Economics of Science and Technology, in
Arrow and the Ascent of Economic Theory New York: New York University Press.
Dasgupta, P. and P. David (1994). Towards a New Economics of Science, Research Policy. 23(5), pp. 487-521.
David, P. (1998). Common Agency Contracting and the Emergence of Open Science Institutions, AEA Papers
and Proceedings, 88(2), pp. 15-21.
Duncan, G. and B. Holmlund (1983). Was Adam Smith Right After All? Another Test of the Theory of
Compensating Differentials, Journal of Labor Economics, 1(4), pp. 366-379.
Ehrenberg, R.G. (1992). The Flow of New Doctorates, Journal of Economic Literature, 30(2), pp. 830-75.
Entorf, H. and F. Kramarz (1998). The Impact of New Technologies on Wages: Lessons from Matching Panels on
Matching Employees and their Firms, Economics of Innovation and New Technology, 5(2-4), pp. 165-198.
Feldman, M.P. (1994), An Examination of the Geography of Innovation, Industrial and Corporate Change, 2, pp.
312-333.
Fleming, L. and O. Sorenson (2003). Science as a Map in Technological Search, mimeo, Harvard Business
School.
Foray, D. (2003). The Economics of Knowledge. Cambridge: MIT Press.
Friedman, M. and S. Kuznets (1954). Income from Independent Professional Practice. New York: National Bureau
of Economic Research.

28

Freeman, R. (1975). Supply and Salary Adjustments to the Changing Science Manpower Market: Physics, 19481973, American Economic Review 65(1), pp. 27-39.
Gambardella, A. (1995). Science and Innovation: The U.S. Pharmaceutical Industry During the 1980s. Cambridge
University Press.
Gibbons, R. and L. Katz (1992). Does Unmeasured Ability Explain Inter-Industry Wage Differentials? Review of
Economic Studies 59, pp. 515-35.
Gittelman, M. and B. Kogut (2003). Does Good Science Lead to Valuable Knowledge? Biotechnology Firms and
the Evolutionary Logic of Citation Patterns, Management Science, 49(4), pp. 366-382.
Hamermesh, D. (1977). Economic Aspects of Job Satisfaction, Essays in Labor Market Analysis edited by O.
Ashenfelter and W. Oates, New York.
Hamermesh, D. (1999). Changing Inequality in Markets for Workplace Amenities, Quarterly Journal of
Economics, 114(4), pp. 1085-1123.
Hamermesh, D. and S. Oster (2002). Tools or Toys?: The Impact of High Technology on Scholarly Productivity,
Economic Inquiry. 40(4), pp. 539-555.
Hamermesh, D. and A. Rees (1984). The Economics of Work and Pay. New York.
Hamilton, L. (1992). Regression with Graphics. Brooks/Cole Publishing.
Henderson, R. and I. Cockburn (1994). Measuring Competence? Exploring Firm Effects in Drug Discovery,
Strategic Management Journal Winter Special Issue, 15, pp. 63-84.
Henderson, R. and I. Cockburn (2001). Publicly Funded Science and the Productivity of the Pharmaceutical
Industry. in A. Jaffe, J. Lerner, and S. Stern (eds.) Innovation Policy and the Economy, Cambridge: MIT
Press.
Hicks, D. (1995). Published Papers, Tacit Competencies and Corporate Management of the Public/Private
Character of Knowledge, Industrial and Corporate Change 4(2), pp. 401-24.
Hsu, D. (2003). What do Entrepreneurs Pay for Venture Capital Affiliation? mimeo, University of Pennsylvania.
Hwang, H., Reed, W.R. and C. Hubbard (1992). Compensating Wage Differentials and Unobserved Productivity,
Journal of Political Economy, 100(4), pp. 835-858.
Hwang, H., Mortensen, D. and W.R. Reed (1998). Hedonic Wages and Labor Market Search, Journal of Labor
Economics, 16(4), pp. 815-847.
Kostiuk, P.F. (1990). Compensating Differentials for Shift Work, Journal of Political Economy, 98(5), pp. 10541075.
Kreps, D. (1997). Intrinsic Motivation and Extrinsic Incentives, American Economic Review, 87, pp. 359-64.
Kremer, M. (1998). Patent Buyouts: A Mechanism for Encouraging Innovation, Quarterly Journal of Economics,
113(4), pp. 1137-67.
Kuhn, T. (1962). The Structure of Scientific Revolutions Chicago: University of Chicago Press.
Lamb, D. and S.M. Easton (1985). Multiple Discovery: The Pattern of Scientific Progress Averbury Publishing.
Latour, B. and S. Woolgar (1986). Laboratory Life: The Construction of Scientific Facts, Princeton: Princeton
University Press.
Lerner, J. and J. Tirole (2002). Some Simple Economics of Open Source, Journal of Industrial Economics, 52, pp.
197-234.
Levin, R., Klevorick, A., Nelson, R., and S. Winter (1987). Appropriating the Returns from Industrial Research
and Development, Brookings Papers on Economic Activity, 3, pp. 783-831.
Levin, S.G. and P. Stephan (1991). Research Productivity over the Life Cycle: Evidence for Academic Scientists,
American Economic Review, 81(1), pp. 114-32.
Lim, K. (2000). Do Firms Have to Perform Basic Research in order to Capture Spillovers, mimeo, MIT Sloan
School.
Merton, R. (1957). Priorities in Scientific Discovery: A Chapter in the Sociology of Science, American
Sociological Review, 22(6), pp. 635-59.
Merton, R. (1973). The Sociology of Science: Theoretical and Empirical Investigation, Chicago: University of
Chicago Press.
Miller, A. (1999). Do Minority MBA Candidates Pay for Diversity Programs, mimeo, MIT Department of
Economics.
Mortensen, D.T. (1990). Equilibrium Wage Distributions: A Synthesis, Panel Data and Labor Market Studies,
edited by J. Hartog, G. Ridder, and J. Theeuwes, Amsterdam: North-Holland.
Murphy, K. and R. Topel (1987). Unemployment, Risk, and Earnings: Testing for Equalizing Differences in the

29

Labor Market, Unemployment and the Structure of Labor Markets Basil Blackwell.
Murray, F. (2003). Innovation as Co-Evolution of Scientific and Technological Networks, Research Policy. 31(89), pp. 1389-1403.
Nelson, R. (1959). The Simple Economics of Basic Scientific Research, Journal of Political Economy, 67(2), pp.
297-306.
Nelson, R. and R. Rosenberg (1994). American Universities and Technical Advance in Industry, Research Policy,
(23)3, pp. 323-348.
Owen-Smith, J. and W. Powell (2001). Careers and Contradictions: Faculty Responses to the Transformation of
Knowledge and its Uses in the Life Sciences, Research in the Sociology of Work, V. 10.
Powell, W., Koput, K., and L. Smith-Doerr (1996), Interorganizational Collaboration and the Locus of Innovation:
Networks of Learning in Biotechnology, Administrative Science Quarterly, 41, pp. 116-145.
Price, D.J. de Solla, and Beaver, D.D. (1966). Collaboration in an Invisible College, American Psychologist, 21,
pp. 1011-18.
Romer, P. (1990). Endogenous Technological Change, Journal of Political Economy, 98, pp. S71-S102.
Rosen, S. (1974). Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition, Journal of
Political Economy, 82(1), pp. 34-55.
Rosen, S. (1986). The Theory of Equalizing Differentials, Handbook of Labor Economics, Vol. 1, ed. O.
Ashenfelter and R. Layard, pp. 641-692.
Rosenberg, N. (1974). Science, Invention, and Economic Growth, Economic Journal, 84(333), pp. 90-108.
Rosenberg, N. (1982). Inside the Black Box. Cambridge: Cambridge University Press.
Rosenberg, N. (1990). Why Do Firms Do Basic Research (with Their Own Money)? Research Policy, 19(2), pp.
165-174.
Rosenkopf, L. and Tushman, M.L. (1998). The Coevolution of Community Networks and Technology: Lessons
from the Flight Simulation Industry, Industrial and Corporate Change, 7, pp. 311-346.
Royalty, A.B. (1999). A Discrete Choice Approach to Estimating Workers Marginal Valuation of Fringe
Benefits, Stanford University Department of Economics Working Paper 98-008.
Sattinger, M. (1977). Compensating Wage Differences, Journal of Economic Theory, 16(2), pp. 496-503.
Scott Morton, F., and J. Poldony (1999). Love or Money?: The Effect of Owner Motivation in the California Wine
Industry NBER WP 6743.
Stephan, P. (1996). The Economics of Science, Journal of Economic Literature, 34, pp. 1199-1235.
Stephan, P. and S. Levin (1992). Striking the Mother Lode in Science: The Importance of Age, Place and Time.
Oxford University Press.
Stern, S. (1995). Incentives and Focus in University and Industrial Research: the Case of Synthetic Insulin, The
University-Industry Interface and Medical Innovation edited by A. Geligns and N. Rosenberg, Institute of
Medicine, National Academy Press.
van Reenen, J. (1996). The Creation and Capture of Rents: Wages and Innovation in a Panel of U.K. Companies,
Quarterly Journal of Economics, 111(1), pp. 195-226.
Viscusi, W.K. 1993). The Value of Risk to Life and Health, Journal of Economic Literature, 31, pp. 1912-1946.
Watson, J. (1968). The Double Helix: A Personal Account of the Discovery of the Structure of DNA.
Weisbrod, B. (1983). Nonprofit and Proprietary Sector Behavior: Wage Differentials and Lawyers, Journal of
Labor Economics 3, pp. 246-263.
Weiss, Y. (1976). The Wealth Effect in Occupational Choice, International Economic Review, 17(2), pp. 292-307.
Wright, B. (1983). The Economics of Invention Incentives: Patents, Prizes, and Research Contracts, American
Economic Review, 73(4), pp. 691-707.
Zucker, L., and M. Darby (1996). Star Scientists and Institutional Transformation: Patterns of Invention and
Innovation in the Formation of the Biotechnology Industry, Proceedings of the National Academy of
Science, 93(23), pp. 709-716.
Zucker, L., Darby, M., and M. Brewer (1998). Intellectual Human Capital and the Birth of U.S. Biotechnology
Enterprises, American Economic Review, 88(1), pp. 290-306.
Zuckerman, H. and R. Merton (1971). Patterns of Evaluation in Science: Institutionalization, Structure, and
Functions of the Referee System, Minerva, 9, pp. 66-100.

30

TABLE 1
DEFINITIONS AND DESCRIPTIVE STATISTICS
VARIABLE

DEFINITION

MEAN

JOB MARKET EXPERIENCE

# OFFERS RECVD Number of offers received

164

2.88

1.00

ACCEPTED JOB

164

0.30

0.49

Accepted this job = 1, No = 0

JOB OFFER CARDINAL RECORD INFORMATION

JOB TYPE
1 = Established Firm, 2 = Startup Firm, 3 = Government,
4 = Medical School/Center, 5 = University, 6 = Postdoc
Monetary Compensation and Career Incentive Measures
SALARY

Annual starting salary (in US Dollars)

STOCK_DUMMY

Job offer includes stock options = 1, No = 0

PROMOTION

Likert Scale rating (1-5) of opportunities for internal promotion

164

STD.
DEV.

See Appendix

121

62263.95

31553.04

72

0.36

0.48

111

3.49

1.29

114

0.92

0.27

104

3.89

1.12

111

0.46

0.50

99

0.00

0.57

112

4.07

0.86

134

1.96

0.99

124

1.90

0.99

116

1.90

0.98

117

1.93

1.00

Scientific Orientation Indicators

PERMIT_PUB

Permission to publish in external journals = 1, No = 0

INCENT_PUB

Likert Scale rating (1-5) of incentives to publish in refereed

outside journals
Job allows continuation of current research project = 1, No = 0

CONTINUE
RESEARCH
SCIENCE
INDEX
EQUIPMENT

First principal factor of PERMIT_PUB, INCENT_PUB and

CONTINUE RESEARCH; see Hamilton (1992)
Likert Scale rating (1-5) of access to cutting-edge equipment

JOB OFFER ORDINAL RECORD DATA (1 =highest)

MONETARY
Ranking of offer in terms of monetary compensation
RESEARCH
QUALITY
FLEXIBILITY

Ranking of offer in terms of internal research environment

FUNDING

Ranking of offer in terms of flexibility to choose research

projects
Ranking of offer in terms of availability of research funding

CAREER

Ranking of offer in terms of impact on career advancement

130

1.95

0.99

JOBFIT

Ranking of offer in terms of how well it fits with prior

research experience

116

1.91

1.01

31

TABLE 2
NON-PARAMETRIC COMPARISON OF DEVIATIONS OF SALARY MEANS
BY SCIENCE ATTRIBUTES
PERMIT_PUB

CONTINUE
RESEARCH
(0 v. 1)

(0 v. 1)

INCENT_PUB
({1, 2, 3} v. {4, 5})

Overall Sample
Average Difference

14200

16809

6694

(7241)

(5189)

(2678)

t-stat for means equality

1.961

3.239

2.500

Degrees of Freedom

26

30

p-value

0.086

0.003

0.018

7200

8143

8430

(2352)

(5954)

(3223)

t-stat for means equality

3.061

1.368

2.615

Degrees of Freedom

12

18

p-value

0.016

0.197

0.0.18

Non-Academic Offers Only

Average Difference

Notes: Each average difference cell contains the difference in the average deviation in salary for a science characteristic, using
the following procedure. For each science characteristic, we identified those individuals whose offers differed in terms of
that characteristic. For these individuals, we then computed (a) the average salary offer for that individual by that specific
characteristic and (b) the (weighted) average salary offer for each individual, where the weight associated with each value
of the characteristic is equal to 0.5. Finally, to abstract away from differences in the level of salary across individuals, we
subtracted (b) from (a). We then performed a t-test for the equality of the means based on these average salary deviations
by scientific characteristic.
Though INCENT_PUB is a five-point Likert scale measure, we implemented the above procedure by grouping the
INCENT_PUB responses into two groups ({1, 2, 3} versus {4, 5}).

32

TABLE 3
HEDONIC WAGE REGRESSION: OVERALL SAMPLE
DEPENDENT VARIABLE = LN(SALARY), # of Observations = 121
PERMISSION TO PUBLISH

PERMIT_PUB

(3-1)
Baseline
(NO FE)

(3-2)
Baseline
(w/ FE)

0.027
(0.186)

-0.266
(0.114)

(3-3)
Full Model
(w/ FE)
-0.191
(0.105)

CONTINUE
RESEARCH
INCENT_PUB

COMBINATION
MODEL
(3-4)
Full Model
(w/ FE)

(3-5)
Full Model
(w/ FE)

(3-6)
Full Model
(w/ FE)

0.063
(0.033)

-0.114
(0.053)
0.057
(0.030)

-0.078
(0.057)
0.053
(0.031)

0.046
(0.021)
0.234
(0.074)
0.002
(0.043)
NO

0.042
(0.021)
0.260
(0.067)
-0.0001
(0.043)
NO

YES
(52; Sig.)
0.958

YES
(52; Sig.)
0.954

0.031
(0.023)
0.190
(0.077)
-0.002
(0.044)
YES
(5)
YES
(52; Sig.)
0.958

-0.089
(0.103)
-0.134
(0.060)
-0.036
(0.028)

SCIENCE INDEX
EQUIPMENT

SCIENCE INDEX MODEL

CONTROLS

PROMOTION
STOCK_DUMMY
ACCEPTED JOB
JOBTYPE
CONTROLS
INDIVDUAL FIXED
EFFECTS
R-squared

NO

NO

NO

YES
(52; Sig.)
0.915

0.001

0.041
(0.025)
0.196
(0.085)
-0.013
(0.040)
YES
(5; Sig.)
YES
(52; Sig.)
0.955

Notes: Only persons with multiple job offers are included.

Standard errors are shown in parenthesis; significant coefficients (10%) are shown in bold.
Sig. stands for joint significance of fixed effects or job type controls (at 10% level)

33

TABLE 4
HEDONIC WAGE REGRESSION: NON-ACADEMIC OFFERS
DEPENDENT VARIABLE = LN(SALARY), # of Observations = 71
PERMISSION TO
PUBLISH
(4-1)
PERMIT_PUB

FACTOR MODEL
(4-2)

-0.150
(0.077)
-0.109
(0.047)
-0.015
(0.038)

SCIENCE INDEX
EQUIPMENT
CONTROLS
PROMOTION
STOCK_DUMMY
JOB ACCEPTED
INDIVDUAL
FIXED EFFECTS
R-squared
Notes:

0.056
(0.029)
0.092
(0.066)
-0.049
(0.047)
YES
(30; Sig.)
0.967

0.054
(0.029)
0.105
(0.065)
-0.021
(0.048)
YES
(30; Sig.)
0.970

Only persons with multiple job offers are included.

Regressions exclude postdoctoral positions and job offers from universities.
Standard errors are shown in parenthesis; significant coefficients (10%) are shown in bold.
Sig. stands for joint significance of fixed effects (at 10% level).

34

TABLE 5
REGRESSION: JOB OFFER COMPARISON RANKINGS
DEPENDENT VARIABLE: MONETARY
(5-1)

(5-2)

All Job Types

Sample

RESEARCH
QUALITY
FLEXIBILITY

-0.34

JOBFIT

0.20
(0.12)
0.23
(0.12)
0.34
(0.13)

CAREER
FUNDING

Exclude
Academic
Job Offers
-0.32
(0.16)

(5-3)

(5-4)

All Job Types

Exclude
Academic
Job Offers

0.29
(0.16)
-0.04
(0.16)
0.26
(0.17)

-0.39
(0.13)
0.32
(0.12)
0.30
(0.13)
0.17
(0.12)

-0.27
(0.19)
0.35
(0.17)
0.04
(0.18)
0.18
(0.17)

0.48

0.41

0.48

0.38

134

74

134

74

INDIVIDUAL FIXED
EFFECTS

R-squared
# of Observations
Notes:

Only persons with multiple job offers are included.

Regressions 5-2 and 5-4 exclude postdoctoral positions and job offers from universities.
Standard errors are shown in parenthesis; significant coefficients (10%) are shown in bold.

35

TABLE 6A
COMPARISON OF SINGLE AND MULTIPLE JOB OFFERS
SINGLE OFFERS
SALARY

58074.01
(26859.57)
0.96
(0.19)
0.28
(0.46)
3.67
(1.24)
3.71
(1.38)

PERMIT_PUB
CONTINUE RESEARCH
INCENT_PUB
EQUIPMENT
Notes:

MULTIPLE OFFERS
61958.85
(31394.96)
0.91
(0.28)
0.46
(0.50)
3.88
(1.13)
4.06
(0.86)

Number of observations is different for every cell, depending on number of missing values.
Standard errors are shown in parenthesis.

TABLE 6B
COMPARISON OF ACCEPTED AND REJECTED OFFER CHARACTERISTICS
REJECTED OFFERS
SALARY

60925.06
(32647.24)
0.91
(0.29)
0.48
(0.50)
3.93
(1.08)
3.82
(0.91)

PERMIT_PUB
CONTINUE RESEARCH
INCENT_PUB
EQUIPMENT
Notes:

ACCEPTED OFFERS
64782.93
(29602.07)
0.95
(0.22)
0.46
(0.50)
3.93
(1.19)
4.43
(0.66)

Number of observations is different for every cell, depending on number of missing values.
Standard errors are shown in parenthesis.

36

APPENDIX A
TABLE A1. CORRELATION OF JOB OFFER RECORD DATA

SALARY
SALARY

PERMIT_
PUB

CONTINUE
RESEARCH

INCENT_
PUB

EQUIPMENT

STOCK_
DUMMY

PROMOTION

1.0000

PERMIT_PUB
CONTINUE
RESEARCH
INCENT_PUB

-0.0082
-0.0728

1.0000
0.0499

1.0000

-0.2729

0.3204

0.2933

1.0000

EQUIPMENT

0.1047

-0.0318

-0.0398

0.1787

1.0000

STOCK_
DUMMY
PROMOTION

0.4680

-0.0757

-0.3264

-0.3953

0.0932

1.0000

0.2173

-0.0231

0.2079

0.0422

0.0532

0.2742

1.0000

Note: Significant (5%) coefficients are shown in bold.

TABLE A2. WAGE REGRESSION: ACADEMIC VERSUS PRIVATE-SECTOR JOBS

DEPENDENT VARIABLE = LN(SALARY), # of Observations = 121
JOBTYPE
(Baseline = Established Firm)

Established Firm

Baseline
(w/o Person FE)

No of
Offers
40

Startup Firm

26

Government

Medical Center

37

University

34

Postdoc

21

-0.17
(0.12)
-0.19
(0.23)
-0.17
(0.12)
-0.41
(0.13)
-0.91
(0.13)
NO

INDIVDIUAL FIXED EFFECTS

R-squared
Notes:

0.34

Baseline
(w/ Person FE)

0.08
(0.08)
0.04
(0.16)
-0.19
(0.08)
-0.40
(0.10)
-0.40
(0.10)
YES
(52; Sig.)
0.93

Only persons with multiple job offers are included.

Standard errors are shown in parenthesis; significant coefficients (10%) are shown in bold.
Sig. stands for joint significance of fixed effects (at 10% level).

37

TABLE A3. CROSS TABULATIONS: JOB OFFER COMPARISON RANKINGS

(Rank = 1 is highest)

MONETARY
(Ranking of
monetary
compensation)

Table A3c: Money and Research Environment

MONETARY
RESEARCH QUALITY
(Ranking of
(Ranking of offer in terms of
monetary
internal research environment)
compensation)
1
2
3
4
5

Table A3a: Money and Jobfit

JOB FIT
(Ranking of offer in terms of how well it fits
with prior research experience)
1

24

16

19

20

21

19

26

15

Table A3b: Money and Career Advancement

MONETARY
CAREER
(Ranking of
(Ranking of impact on career advancement)
monetary
compensation)
1
2
3
4
5
6

Table A3d: Money and Flexibility

MONETARY
FLEXIBILITY
(Ranking of
(Ranking of offer in terms of flexibility to
monetary
choose research projects)
compensation)
1
2
3
4
5
6

26

20

16

19

20

25

25

17

13

5
6

38

1
1

FIGURE 1

Distribution of Individual Fixed Effects Coefficients in Regression (3-3)

1
0.8
0.6
0.4
0.2
0
1

11

13

15 17

19

21 23

25

27 29

31

33 35

-0.2
-0.4
-0.6
-0.8
-1
-1.2
-1.4

By Ascending Rank of Individual Fixed Effect

Number of Individuals = 52

39

37

39

41 43

45

47 49

51